Human StarCraft II e-athletes crushed by neural net ace – DeepMind's AlphaStar

AlphaStar, DeepMind’s latest AI bot, crushed professional gamers playing popular strategy video game StarCraft II during a demonstration broadcast online Thursday. DeepMind has had its eyes on the StarCraft II for a while. Back in 2016, it announced a partnership with Starcraft creators Blizzard Entertainment, to work together …

[Joe W]

Yeah, right...

"Demis Hassabis said he believed he code could be used to help predict the weather"

Well, maybe as the second layer under a proper dynamic (i.e. deterministic atmospheric circulation model), doing some sort of statistical downscaling. Though in my experience the linear models are really hard to beat.

[find users who cut cat tail]

Re: Yeah, right...

Although we can still improve the computational part (and gather better observational data), predicting weather is hard because of positive Lyapunov exponents galore. No AI will fix that.

[Charles 9]

Re: Yeah, right...

Are they irreducibly positive and/or weather prediction provably chaotic?

[Antonius_Prime]

It's clever, but it's limited.

No fog of war, same map all the time, same races all the time.

It's been trained to do one map and race very, very, very well. So of course it will excel at that. It's been thought by rote!

If they can train the bot to fling out toxic speech too, they'll essentially have created a teenage boy doing state exams! (A Levels, Leaving Cert, Matric, SAT's etc)

[Anonymous Coward]

Re: It's clever, but it's limited.

"No fog of war, same map all the time, same races all the time.

It's been trained to do one map and race very, very, very well. So of course it will excel at that. It's been thought by rote!"

True that. I think any bad player spending 10 years training on the same map with no fog of war can flat out crush any pro player. It doesn't really show anything ...

[Prosthetic Conscience]

Re: It's clever, but it's limited.

Not true, there is fog of war that was one of the main things, the advantage it had was that with this SC2 binary they use it's minimap is in a way its main input but with all the detail of looking at the screen. It's still a massive advantage as that allows it to use that 'realistic' APM on really really useful actions indeed.

The one we should be looking at is the version that was defeated by MaNa, that one does look at the screen the same way and it was a lot more natural and the micro was still off the hook.

The interesting thing is how it applies a strategy that's very unconventional and I think that combined with the inhuman micro skills that will always defeat human players (but lets see some Koreans try too). as humans have so far played only against humans and don't have 200 years worth of playing (as AlphaStar did)

[Antonius_Prime]

Re: It's clever, but it's limited.

See, you've info there I hadn't gleaned from the article.

Thanks!

My main point is that it has learned by rote, only its memory is a damn good way better than a 15-18 year old males! It obviously learns, Borg-like, from strategies that are successful.

And do we really want to unleash an AI on the Koreans? Poor thing won't stand a chance!

[Charlie Clark]

Re: It's clever, but it's limited.

inhuman micro skills that will always defeat human players (but lets see some Koreans try too)

It should be possible to limit those to an arbitrary speed, though the current limitation is probably close to that of a human: see something, process and react.

[amanfromMars 1]

Actually, Mere Mortals are Lost in the CyberSpace Sector and Virtually Real Worlds

But, all was not lost for the humans. In the single live match streamed during the broadcast, MaNa managed to convincingly beat AlphaStar despite a strong opening from the agent. The commentators said during its matches with TLO AlphaStar seemed to play more daringly, carrying out wackier strategies not usually seen in normal human play . When it played MaNa, however, it seemed to settle down into a more conventional style of play and lost.

And that makes humans Catastrophically Vulnerable and XSS Streamly Susceptible to Alphabet Spaghetti/Mega MetaData Base Search Engine Feeds and Seeds copying Extreme Behaviour .......with Realms in Mad Genius and Domains in Genius Madness to Exercise?

And whilst it may be considered practically new, it has never been unknown to a very selective and easily highly disruptive few? ....... Alternative Operations in Advanced IntelAIgent Missions

And can you imagine how much further and deeper the travels in such fields in three years and what/who one encounters on Parallel Courses? :-)

Subject: An Engaging Competition or Emergent Opposition, or Both and Something Else Too and Beautifully Creative

Date: 30 January 2016 at 09:46:55 GMT

To: [e-mail]

Hi, Google DeepMinded,

Not so much a CV, more an AIMission Statement with tried and tested roadmaps to/from Remote Virtual Command and Control of Earthed SCADA Systems ……. http://forums.theregister.co.uk/forum/1/2016/01/29/ai_in_tv_film_books_games/#c_2762789

Search Engine Optimisation v2.0 [and above] is surely logically a Future Product Placements Engine …… Advanced Intelligence Resource with Immaculate Source, with the likes of a Google not searching for answers, both popular and controversial, but providing them with streams of supporting evidence.

Such would be akin to the Private Mentoring with Pirated Monitoring of Future Events with AIDerivative Programming for Projects/Semi-Autonomous, Self-Actualisation of Virtual Realities.

It is difficult, and maybe even impossible, to see or imagine a defence against such in an attacking configuration.

Regards,

The Defence/Prosecution rests, M'Lud ...... and Encourages Just Deliberation and Enlightened Sentencing Displaying Advanced Intelligence rather than Supporting Arrogant Ignorance.

[Hemmels]

"Imitation" learning for AI

I much prefer "imitation" learning over training by "playing against each other" ala Dota2's OpenAI.

I cringe a bit when I think about how many CPU cycles are being used for this sort of stuff, slowly heating our planet in the process.

[tmTM]

Re: "I cringe a bit when I think about how many CPU cycles are being used for this sort of stuff

Just think of the power being used to generate increasingly worthless crypto coins

[druck]

Not playing the same game

It receives a raw interface describing the current state of the game with a list of available units and their properties as input

AlphaStar has a few advantages compared to its human opponents. It played a slightly dated version of the game “geared for research” rather than for entertainment, and could glance at the whole map. Everything not hidden by the fog of war, including its own units and its enemies can be seen at once. Humans don’t get this luxury, and instead only see bits and pieces by clicking on parts of the map.

It would have to be pretty shoody AI not to win with such an advantage.

When the AI is playing the same game, i.e. with the same interface as the human, and it wins, then report the story.

[amanfromMars 1]

Re: Not playing the same game

When the AI is playing the same game, i.e. with the same interface as the human, and it wins, then report the story. .... druck

Would you be able to easily recognise an AI playing humans on an interface not entirely dissimilar to the one we be testing here, druck? And would it being a difficulty, create problems for humans.

But hey, that is progress. Better get used to it, for it is certainly here to stay to move things on into Other Greater Ages. What humans are resolved to do that easily and stealthily with the multiple tools and myriad programs readily available for use?

Some might already realise the Fundamental Shift and Irrevocable Change in the Ways Things be Done and Realised and would appear to fear and have doubts and worries about it .......

"We face significant changes in the domestic and global environment; we must be ready to meet 21st century challenges and to recognize emerging threats and opportunities," the report reads.

"To navigate today’s turbulent and complex strategic environment, we must do things differently." .... https://www.theregister.co.uk/2019/01/26/security_roundup_250119/

[LuDo]

Re: Not playing the same game

Does here seems AI talking is?

[dtl]

Not fair

I'm not saying I'm not impressed, but doing a bit of machine learning research myself I can see how this is possible. However, StarCraft is intentionally hard for human players to control. You have to remember a ton of things, both long term and short term, you have to manage your attention and you have to train physically like a piano player to translate your actions into keyboard/mouse clicks.

The combination of these leads to a steadily growing learning curve the more time you put into the game, which is of course exactly what the game is designed for.

Now you want to test if your neural network (I refuse to call it AI, it's nowhere even close to that) can do strategy, so you give it a much simplified interface to work with. That's OK for this part of research, and it's also OK to pit it against human players to see how it performs, but don't interpret too much in the results. We will talk again when this neural net can move under it's own power in front of a screen and a keyboard and play the game under the same conditions as a human pro player.

Until then, we will have to learn how much processing power the bot needs after training (i.e. to just execute the neural net). This is usually far less then required for training the network, so if some better bots come out for us humans to train against, I wouldn't mind, especially if you can select the version of it. For example you have trouble defending against a certain strategy and can select a bot which implements it well.

[Anonymous Coward]

Pfft, I'm not concerned about Alphastar

Let's see how it deals with me as its Terran team-mate, when I mistakenly drop a nuke on its own advancing troops!! Any REAL Starcraft player would know how to deal with that.

Nuclear launch detected!

[Johnny Canuck]

I watched it.

It strangely had a problem with repeated drops in its home base. Alphastar would start moving out and MaNa would drop 2 Immortals in its mineral line. Then Alphastar would bring its army back and MaNa would retreat. Rinse, repeat. This allowed MaNa enough time to build a strong army and eventually roll over Alphastar.

[BinkyTheMagicPaperclip]

Re: I watched it.

Sounds like a computer to me. Reacts based on historic patterns, does not think. Whereas a human player might get a little suspicious after it happens a few times and try something different, or risky.

[P.B. Lecavalier]

professional

I don't know which one is the most depressing. Is it that a machine beats humans, or that there are humans whose job is to be a "professional SC gamer"?

[Charles 9]

Re: professional

Well then, how does the phrase Major League Gaming depress you?

[Persona]

Re: professional

Hmmmmm …….. I had to Google it.

[KillStuffMount]

Re: professional

Entertainment sells. If you have an audience, you can make money off it by selling advertising/tickets and so on, which means prize money.

Half the world enjoys watching 22 men kick an inflated sheep's bladder around a rectangle of grass for 90 minutes at a time, there's quite the industry built up around that too :)

[9Rune5]

Re: professional

Half the world enjoys watching 22 men kick

Well. Two wrongs don't make a right.

[Charles 9]

Re: professional

Sure it can. Majority rule. Especially when you throw in the people who watch 22-24 men run around with, kick, and throw an oblate spheroid around a rectangular field for anywhere from 60-80 minutes at a stretch, not to mention various popular local variations like the one on an elliptical field that seems to draw the most people at a time to a single sports venue every year.

[Anonymous Coward]

Re: professional

There's even enough money involved in the "22 guys kicking a sheep's bladder" thing to (still) encourage a lot of international corruption!